Newsgroups: comp.ai.genetic
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!howland.reston.ans.net!swrinde!pipex!sunic!news.chalmers.se!news.gu.se!gd-news!d6201.shv.hb.se!sa209
From: sa209@utb.shv.hb.se (Claes Andersson)
Subject: Re: GP with a REALLY small pop
Message-ID: <sa209.91@utb.shv.hb.se>
Sender: usenet@gdunix.gd.chalmers.se (USENET News System)
Nntp-Posting-Host: d6201.shv.hb.se
Organization: Department of Scocial Science
References: <D0Lq5p.GI9@aisb.ed.ac.uk>
Date: Mon, 12 Dec 1994 17:53:57 GMT
Lines: 39

In article <D0Lq5p.GI9@aisb.ed.ac.uk> roberts@aisb.ed.ac.uk (Robert Saunders) writes:

>Hi all

>I'm currently trying to do a final year project using Genetic Programming
>to interactively evolve 3D objects based on a user's ideas of aestetics.
>One of the major problems with this approach is that the population has
>to be very small (9-25) so that the user can evaluate all the members.
>I have been trying to think of ways to use the information gained from
>the user work for the evolution on the most effective way.

>One idea I have had which I would like some feedback on is to keep members
>of previous populations beyond their normal lifespan in GP and reduce their
>fitness according to age.

>This would allow a more continuous model of evolution, allowing older
>members a smaller (but significant) chance to contribute to later
>populations than would be normal. Really old members could be killed
>off once their fitness had fallen to a very low level but I doubt it
>would be neccessary because of space considerations because of the
>small population sizes.

>It would seem to have some parallels in nature and I wondered if
>anyone had heard of any research into this type of thing.

>Cheers

>Rob

 Hello Rob,

 As a matter of fact I have had/still have the exactly same problem. I have 
made an education software for evolution theory and have to use a population 
of 12 speciemen. It's not really fun to use crossover in such a small 
population since it converges very fast. The result is that it can serve as 
an educational model but adapts quite poorly if you aren't really "kind" to 
the biomorphs.

Claes Andersson. University of Bors. Sweden
